D-RISK: Validating the autonomous vehicle safety-case using simulation and adaptive test generation

D-RISK：使用模拟和自适应测试生成验证自动驾驶汽车安全案例

• 批准号: 104528
• 金额: $ 380.85万
• 依托单位: DRISK.AI LIMITED
• 依托单位国家: 英国
• 项目类别: Collaborative R&D
• 财政年份: 2019
• 资助国家: 英国
• 起止时间: 2019 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=104528
• 关键词: RISK; Validating; autonomous; vehicle; safety

"**Need.**Simulation will be a vital component in the Connected Autonomous Vehicle (CAV) safety-case, enabling the validation and certification of autonomous Vehicle Control Systems (VCS). This is essential in cities where the density and diversity of other agents, including road-users, pedestrians, infrastructure and signage, as well as intermodal-connectivity, present a limitless set of potential scenarios that must be safely negotiated, either individually or collectively.No existing simulation system enables a full safety-case approach to certification within an overarching risk architecture based on clearly-defined key performance indicators and areas. Limitations on incumbent physical simulators and models fail to capture:* High complexity of edge-case scenarios in these contexts, which are difficult to predict and occur with low frequency compared to the majority of everyday driving scenarios* Unpredictable responses of other road users and pedestrians to CAVs* Rapidly changing technology and regulatory landscapes* Relevant socioeconomic and human factorsFundamentally, too many individual types and levels of risk exist for just one kind of continuous real-world simulation and corresponding simulation platform.**Disruptive solution.**In response, aiPod have brought together an ambitious cross-industry consortium, including Imperial College London, Claytex, DG Cities and Transport for London, to develop and integrate, as part of a simulator system, a novel platform-agnostic scenario generator for simulated testing of SAE Level-4/5 urban autonomy at the scale of the individual sensor, CAV and entire transport network.Applicable to all potential CAV variants and any simulation architecture feeding a safety-case approach, the solution targets a step change in capability to:1. Identify and structure complex realistic edge-case scenarios2. Adaptively formulate the most relevant test routines to virtually-validate CAV decision-making performance3. Qualify and quantify overall risk, aggregated across the relevant scenarios and parameter spaceA multi-level framework targets unprecedented capability to identify complex edge-case test scenarios from multi-mode real and simulated data, human-driven fault-detection and public responses to CAV.Overcoming limitations of existing simulators and models, the project targets a functional architecture that is inherently evolvable and extendable, whilst allowing constraints to test and certify specific deployments. Actionable feedback delivered by the solution and wider project can be leveraged to evolve the overall UK autonomous vehicle simulation capability and regulatory position."

**需要。**仿真将是互联自动驾驶汽车(CAV)安全案例中的一个重要组成部分，使自动车辆控制系统(VCS)能够进行验证和认证。在城市，道路使用者、行人、基础设施和标志等其他主体的密度和多样性以及多式联运连通性提供了一套无限的潜在情景，必须单独或集体安全协商。现有的模拟系统都不支持在基于明确定义的关键绩效指标和领域的总体风险架构内采用完整的安全案例方法进行认证。现有物理模拟器和模型的局限性未能捕捉到：*这些环境中边缘情况场景的高度复杂性，与大多数日常驾驶场景相比，这些场景难以预测且发生频率较低*其他道路使用者和行人对Cavs的不可预测的反应*快速变化的技术和监管环境*相关的社会经济和人为因素基本地说，存在太多的个体类型和风险水平，仅针对一种连续的真实世界模拟和相应的模拟平台。**破坏性的解决方案。**作为回应，aiPod召集了一个雄心勃勃的跨行业联盟，包括帝国理工学院、Claytex、DG Cities和Transport for London，以开发和整合作为模拟器系统的一部分，一个新颖的与平台无关的场景生成器用于在单个传感器、CAV和整个交通网络的规模上模拟测试SAE-4/5级城市自主性。该解决方案适用于所有潜在的CAV变体和任何提供安全案例方法的仿真体系结构，其目标是在以下方面实现阶梯变化：1.识别和构建复杂的现实边缘案例场景2。自适应地制定最相关的测试例程，以虚拟验证CAV决策性能。鉴定和量化在相关场景和参数空间聚集的总体风险一个多级别框架的目标是具有前所未有的能力，从多模式真实和模拟数据、人为故障检测和公众对CAV的响应中识别复杂的边缘用例测试场景。克服现有模拟器和模型的限制，该项目的目标是一个内在可演变和可扩展的功能体系结构，同时允许约束来测试和认证特定部署。该解决方案和更广泛的项目提供的可操作反馈可以被用来发展英国整体的自动驾驶汽车模拟能力和监管地位。“